---
title: "FDA AI/ML SaMD Action Plan and AI Voice Triage in 2026"
description: "FDA's January 2025 lifecycle-management draft guidance, Predetermined Change Control Plans, and the January 2026 deregulatory pivot reshape what AI voice triage looks like for US healthcare in 2026."
canonical: https://callsphere.ai/blog/vw5f-fda-ai-ml-samd-action-plan-voice-triage-2026
category: "AI Strategy"
tags: ["FDA", "SaMD", "PCCP", "AI Voice", "Triage"]
author: "CallSphere Team"
published: 2026-03-21T00:00:00.000Z
updated: 2026-05-07T16:29:55.966Z
---

# FDA AI/ML SaMD Action Plan and AI Voice Triage in 2026

> FDA's January 2025 lifecycle-management draft guidance, Predetermined Change Control Plans, and the January 2026 deregulatory pivot reshape what AI voice triage looks like for US healthcare in 2026.

> The FDA's posture on AI-enabled medical software changed twice in twelve months: a serious lifecycle-management draft in January 2025, then a deregulatory pivot in January 2026. AI voice triage sits in the middle of that swing.

## What the rule says

The FDA's modern AI/ML SaMD framework rests on the January 2021 AI/ML Action Plan, the October 2021 Good Machine Learning Practice (GMLP) guiding principles, the April 2023 Predetermined Change Control Plan (PCCP) draft guidance, and the January 6, 2025 draft guidance on Artificial Intelligence-Enabled Device Software Functions: Lifecycle Management and Marketing Submission Recommendations. The 2025 draft pulls together total product lifecycle expectations: data management, model development, validation, monitoring, and labeling, with explicit attention to performance monitoring after market authorization. PCCPs let a sponsor pre-authorize specified model updates without a new 510(k), provided the change is described, the methodology is documented, and the impact assessment is in place.

On January 6, 2026 the FDA published guidance reducing oversight of certain digital-health and clinical-decision-support software, especially low-risk wearables and CDS that does not direct clinical decisions. As of mid-2025 FDA's public list of AI-enabled medical devices exceeded 1,250.

## What AI voice/chat must do

If an AI voice or chat agent diagnoses, triages, or recommends a medical action, it can be a device function. The FDA's CDS criteria — the "four prongs" — determine whether it falls inside the 21st Century Cures Act CDS exemption: the software must (1) not acquire or process medical images or signals, (2) display information for an HCP, (3) explain its basis so the HCP can independently review, and (4) inform rather than replace HCP judgment. AI triage that talks directly to a patient and recommends action typically does not satisfy prong 2.

For in-scope deployments: build a quality system, document the device description, perform analytical and clinical validation, maintain a model card aligned to GMLP, file a 510(k) or De Novo with a PCCP if planning iterative updates, and stand up post-market performance monitoring. For out-of-scope deployments: write a qualification memo and keep it.

## CallSphere compliance posture

CallSphere positions Healthcare Voice as operations and access — intake, scheduling, refill prompts, benefits, payment, and crisis routing — not as a diagnostic device. The Healthcare Voice Agent's 14 tools enforce that boundary in prompts, tool design, and refusal behavior; clinical questions get routed to humans. The platform is HIPAA and SOC 2 aligned, runs on an encrypted PostgreSQL `healthcare_voice` database, captures full post-call analytics — sentiment (-1.0 to +1.0), lead score (0–100), AI summary, audit trail — and exposes the data customers need for their own SaMD evaluations if they extend functionality. Behavioral-health deployments at [/lp/behavioral-health](/lp/behavioral-health) ship with crisis-routing patterns aligned to FDA expectations for human oversight. The platform runs 37 agents, 90+ tools, 6 verticals, 50+ businesses at 4.8/5. Pricing $149 / $499 / $1,499; [14-day trial](/trial); 22% affiliate. Hub: [/industries/healthcare](/industries/healthcare).

```mermaid
flowchart LR
A[AI Triage Use Case] --> B{Device\nFunction?}
B -- No --> C[Qualification Memo]
B -- Yes --> D[GMLP + Model Card]
D --> E[510k or De Novo]
E --> F[PCCP]
F --> G[Post-Market\nMonitoring]
G --> H[Drift Triggers\nCorrective Action]
```

## Compliance checklist

1. Run a CDS four-prong analysis on every AI voice or chat feature with clinical purpose.
2. Build a written qualification memo for everything that lands out of scope.
3. For in-scope features, document the device description, intended use, indications for use, and contraindications.
4. Implement GMLP guiding principles end to end, especially data quality, training/test separation, and clinically meaningful endpoints.
5. Plan analytical validation (technical performance) and clinical validation (clinical performance) before submission.
6. File a PCCP with the marketing submission if you plan iterative model updates.
7. Stand up post-market performance monitoring with defined drift thresholds and corrective actions.
8. Maintain a model card with version, training window, performance, and known limits.
9. Track real-world performance against pre-specified clinical endpoints.
10. Re-evaluate scope after the January 2026 deregulatory guidance — narrower CDS oversight may move some features out of scope.

## FAQ

**Is AI voice triage always a device?**
Not always. Routing to a queue based on urgency without making a clinical recommendation can stay out of scope. Triage that recommends a clinical action usually is in scope.

**What is a PCCP good for?**
Pre-authorizing iterative model updates so you do not need a new 510(k) for each version.

**Does the 2026 deregulatory guidance remove AI voice from oversight?**
Only for low-risk CDS that meets the four prongs and does not direct clinical decisions.

**Are LLMs themselves devices?**
The LLM by itself is general-purpose software. Embed it in a clinical decision pathway and the resulting function can be a device.

## Sources

- FDA AI/ML SaMD Action Plan (Jan 2021): [https://www.fda.gov/media/145022/download](https://www.fda.gov/media/145022/download)
- FDA Good Machine Learning Practice — Guiding Principles: [https://www.fda.gov/medical-devices/software-medical-device-samd/good-machine-learning-practice-medical-device-development-guiding-principles](https://www.fda.gov/medical-devices/software-medical-device-samd/good-machine-learning-practice-medical-device-development-guiding-principles)
- FDA Draft Guidance — Predetermined Change Control Plans: [https://www.fda.gov/media/166704/download](https://www.fda.gov/media/166704/download)
- FDA AI-Enabled Medical Devices list: [https://www.fda.gov/medical-devices/software-medical-device-samd/artificial-intelligence-enabled-medical-devices](https://www.fda.gov/medical-devices/software-medical-device-samd/artificial-intelligence-enabled-medical-devices)
- FDA Artificial Intelligence in SaMD page: [https://www.fda.gov/medical-devices/software-medical-device-samd/artificial-intelligence-software-medical-device](https://www.fda.gov/medical-devices/software-medical-device-samd/artificial-intelligence-software-medical-device)

---

Source: https://callsphere.ai/blog/vw5f-fda-ai-ml-samd-action-plan-voice-triage-2026